摘要:
为解决淀粉工业中淀粉液化液葡萄糖当量(Dextrose Equivalent,DE)值的化学检测方法繁琐复杂问题,研究近红外光谱(near infrared spectroscopy,NIRS)快速检测淀粉液化液DE值的方法。取88个淀粉液化液作为样本,扫描得到近红外光谱图,分为72个初始样本校正集,10个初始样本预测集,剔除6个问题样品,研究了运用多元散射校正(Multiplicative signal correction,MSC)、标准正态变量变化(Standard normal variate,SNV)、平滑(smoothing)、多项式平滑(Savitzky-Golay卷积平滑,SG)、一阶导数(1st derivative,1D)、二阶导数(2nd derivative,2D)等不同预处理方法和偏最小二乘法(partial least squares,PLS)、主成分回归法(principal component regression,PCR)组合的建模效果。计算分析结果表明:多元散射校正结合一阶导数与偏最小二乘法算法建模的初始模型稳定性和预测能力较好;近红外DE值模型的主成分因子数、交叉验证均方差(RMSEC)、交叉验证决定系数(RC)、预测均方差(RMSEP)、预测决定系数(RP)依次为:3、1.53、0.9723、1.44、0.9746。因此该NIRS检测方法快速简便,可用于淀粉液化液DE值的快速检测。
Abstract:
The chemical detection methods of starch liquefaction liquid’s Dextrose Equivalent value are complicated. In order to solve this problem,fast detection of starch liquefying solution’s DE value by near infrared spectroscopy(NIRS)has been researched. The 88 samples of starch liquidfaction liquid and their NIRS were divided into a calibration set of 72 samples and a prediction set of 10 samples beside 6 suspicious samples. The modeling effects of different pretreatment methods such as Multiplicative signal correction,Standard normal variate,smoothing,Savitzky-Golay,1st derivative,2nd derivative,partial least squares and principal component regression combination were studied. Calculation results showed that the model stability and prediction ability by the first derivative(1D)and partial least squares(PLS)algorithm was a suit method. The Factor,RMSEC,RC,RMSEP,RP of NIRS DE model were:3,1.53,0.9723,1.44,0.9746. Therefore,the NIRS detection method can be used for rapid detection of the DE of starch liquefaction liquid.
本站查看
下载:
